Additional considerations

Check stability at operating temperature once thermal conditions have stabilized. A successful test immediately after a cold boot says nothing about memory behavior an hour later, after the DIMMs, PMIC, case air, and area around the CPU have warmed up.

The SPD Hub sensor is useful as a repeatable reference point, but it measures its own local temperature, not the hottest DRAM die. Some modules have no sensor at all. The trend for one kit and the temperature at the first error therefore matter more than someone else’s universal number.

THERMAL LOOP: RECORD THE ENTIRE CHAIN

WHAT TO RECORD DURING THE TEST

PRACTICAL SEQUENCE

1. COLD BASELINE

Start the selected test and record the initial temperature, VDD/VDDQ, fan speed, and system background load.

2. STEADY-STATE WARM-UP

Do not change conditions until the temperature plateaus; an early pass does not establish stability after warm-up.

3. REAL-WORLD WORST CASE

Repeat the workload in a closed case and, if typical for your use, with the graphics card warmed up.

4. DIAGNOSTIC AIRFLOW

With settings unchanged, direct airflow at the DIMMs. If the errors disappear, this indicates inadequate thermal margin, but it does not replace repeat testing.

The IMC (Integrated Memory Controller) is located in the CPU, and its signals pass through the LGA contacts, motherboard traces, and DIMM slots. A missing channel, a sharp deterioration in training after remounting, or instability on one channel therefore calls first for mechanical diagnosis at JEDEC settings, not additional voltage.

A contact frame changes socket geometry and pressure distribution. It may affect temperature and contact, but it does not improve the IMC’s intrinsic quality or guarantee a higher memory data rate.

SIGNAL PATH

ISOLATE THE PROBLEM BEFORE TRYING TO FIX IT

SAFE REINSTALLATION

1. JEDEC, THEN REMOVE POWER

Save the profile and test at JEDEC. Shut down the PC, switch off the power supply, and unplug the power cord; make sure standby power is gone.

2. INSPECT THE SEATING

Follow the motherboard’s ESD instructions; hold the CPU and DIMMs by the edges. With the motherboard horizontal, place the CPU without sliding it or touching the LGA contacts.

3. FOLLOW ONE MANUAL

Tighten the cooler or frame only in the sequence and to the force specified by its manufacturer; do not transfer torque values from other hardware.

4. RETRAIN AND RETEST

After any remounting, clear saved training and repeat cold-boot, warm-restart, and stress testing.

The BIOS version is part of the profile. An update may change MRC (Memory Reference Code, the memory-initialization code), CPU microcode, Intel ME, hidden Auto values, and Memory Fast Boot. Identical visible timings after a flash therefore do not guarantee the same training state.

For the Z790 Apex Encore, ASUS explicitly associates BIOS 2202 with the Intel IPU 2026.1 update and memory compatibility; ME changes at the same time, and its new version may remain after a BIOS rollback. Such a rollback does not always restore the previous software stack.

Unstable memory is dangerous for more than blue screens: an error can enter an archive, project, file system, or synchronized folder without any obvious warning. An experimental profile must not operate on the only copy of important data.

AIDA64 and Intel MLC let you assess changes in memory bandwidth and latency, but their results do not directly indicate FPS gains, changes in frame time, or archiving speed. The real-world effect depends on the workload, CPU/GPU limit, working-set size, and background load.

THREE LEVELS OF MEASUREMENT

Four modules, additional ranks, and higher electrical capacitance increase the load on the controller and signal path. Choose the configuration according to the required capacity: maximum frequency is a task for a 2-DIMM / 1SPC motherboard, while a 4-DIMM / 2SPC system requires a more moderate target and an exact QVL match.

Rank is determined by module organization, not by capacity alone. Two kits with the same gigabyte capacity can present different loads, so check the part number, rank count, and actual configuration of each DIMM.

LARGE-CAPACITY TUNING SEQUENCE

1. CHOOSE THE CAPACITY IN ADVANCE

Do not buy a second random kit later: even the same part number may receive a different IC revision.

2. CHECK THE QVL

Match the exact part number, module count, capacity, BIOS, and recommended slots.

3. START AT JEDEC

Verify that every channel and module works before enabling XMP or manual voltages.

4. LOWER THE DATA RATE BEFORE ADJUSTING VOLTAGE

If XMP is not repeatable, lower the frequency by one or two steps, then find the minimum voltages.

5. TUNE TIMINGS AFTER FREQUENCY

First stabilize training and booting, then tighten the timings in small groups.

6. VALIDATE EVERY MODE

Several cold boots and warm restarts, a hot memory test, and a normal workload.

To reproduce a profile, save its settings, test conditions and results. The materials must make it possible to reconstruct the test bench a month later and understand changes after a BIOS update, remount, or module replacement.

MINIMUM ENTRY IN THE REVISION LOG

• Date, document revision number, and changed section.
• Previous claim, new claim, and reason for the change.
• Primary source or link to a reproducible test.
• Platform, BIOS, DIMM capacity/rank, and test conditions.
• Whether previously published profiles must be retested.